takler

🚧 takler is under construction.

A task scheduler tool for Numerical Weather Prediction (NWP) workflows.

Chinese documentation (中文文档)：https://takler.readthedocs.io/

Quick Start

Install takler with pip:

pip install takler

1. Define a flow

Create test.py, which defines a flow named test with two shell tasks. The trigger on t2 makes it wait until t1 completes:

import asyncio
import sys
from pathlib import Path

from takler.core import Flow
from takler.server import TaklerServer
from takler.tasks.shell import ShellScriptTask
from takler.visitor import pre_order_travel, PrintVisitor

TAKLER_HOME = Path(__file__).parent


def create_flow():
    flow = Flow("test")
    flow.add_parameter("TAKLER_HOME", str(TAKLER_HOME))

    task1 = flow.add_task(ShellScriptTask("t1"))
    task1.add_parameter("TAKLER_SCRIPT", str(Path(TAKLER_HOME, "test/t1.takler")))

    task2 = flow.add_task(ShellScriptTask("t2"))
    task2.add_parameter("TAKLER_SCRIPT", str(Path(TAKLER_HOME, "test/t2.takler")))
    task2.add_trigger("./t1 == complete")

    return flow


async def run_takler(server):
    await server.start()
    await server.run()


def main():
    flow = create_flow()
    pre_order_travel(flow, PrintVisitor(sys.stdout))

    server = TaklerServer()
    server.bunch.add_flow(flow)
    asyncio.run(run_takler(server))


if __name__ == "__main__":
    main()

2. Write the task scripts

Each ShellScriptTask renders a .takler script (a Jinja2 template) into a job script and runs it. The script calls back to the server with child commands: init when it starts and complete when it finishes. Create test/t1.takler:

#!/bin/bash
set -e

takler-client-py init --host {{ TAKLER_HOST }} --port {{ TAKLER_PORT }} \
    --task-id $$ --node-path {{ TAKLER_NAME }}

echo "hello from t1"
sleep 2

takler-client-py complete --host {{ TAKLER_HOST }} --port {{ TAKLER_PORT }} \
    --node-path {{ TAKLER_NAME }}

and test/t2.takler the same way, with echo "hello from t2" as the payload. TAKLER_HOST, TAKLER_PORT and TAKLER_NAME are parameters takler generates for every task; the client used inside the script can be either takler-client-py (installed with takler) or the Go client takler_client, as long as it is on the job's PATH.

3. Run it

Start the server (it listens on localhost:33083 by default):

python test.py

In another terminal, begin the flow and requeue it so the scheduler picks it up:

takler-client-py begin test      # begin acts on a flow name, not a path
takler-client-py requeue /test
takler-client-py show

t1 runs first; once it reports complete, the trigger releases t2, and finally the whole flow reaches complete:

|- test [complete]
  |- t1 [complete]
  |- t2 [complete]

For events, meters, time dependencies, families and more, see the documentation (中文文档).

Documentation

Build documentation locally

Documentation dependencies (Sphinx, pydata-sphinx-theme, etc.) live in the docs dependency group in pyproject.toml, kept separate from the dev group so a docs build doesn't also pull in the linter or test tooling. Build the HTML docs with:

uv run --group docs sphinx-build -b html doc/source doc/build/html

or, using the Sphinx Makefile:

cd doc
uv run --project .. --group docs make html

The generated HTML is written to doc/build/html/index.html. This is the same uv sync --group docs toolchain that Read the Docs uses (see .readthedocs.yml).

LICENSE

Copyright © 2022-2025, developers at cemc-oper.

takler is licensed under Apache License, Version 2.0

takler